Research map / Image generation / Autoregressive and latent generation
MambaVoiceCloning: Efficient and Expressive Text-to-Speech via State-Space Modeling and Diffusion Control
Sahil Kumar, Namrataben Patel, Honggang Wang, Youshan Zhang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add AdaLN style modulation to gated fusion (ablation) | MVC text encoder · LJSpeech long-form | MOS-long: 4.02 -> 4.16 | Improved |
| add gating to AdaLN-only fusion (ablation) | MVC text encoder · LJSpeech long-form | MOS-long: 3.95 -> 4.16 | Improved |
| add gating and AdaLN to concat fusion (ablation) | MVC text encoder · LJSpeech long-form | pitch RMSE (Hz): 2.89 -> 1.92 | Improved |
| add gating and AdaLN to concat fusion (ablation) | MVC text encoder · LJSpeech long-form | MOS-long: 3.64 -> 4.16 | Improved |
| add Expressive Mamba predictor (ablation) | MVC · OOD set | CMOS-N: -0.41 drop when removed | Improved |
| add Bi-Mamba text encoder (ablation) | MVC · OOD set | CMOS-N: -0.38 drop when removed | Improved |
| add Temporal Bi-Mamba encoder (ablation) | MVC · OOD set | CMOS-N: -0.36 drop when removed | Improved |
| replace BiLSTM text encoder with 6 Mamba layers (ablation) | MVC text encoder · LJSpeech | MOS ID: 3.61 -> 3.87 | Improved |
| increase Mamba text encoder depth from 6 to 7 layers (ablation) | MVC text encoder · LJSpeech | MOS ID: 3.87 -> 3.90 | No clear effect |
| increase Mamba text encoder depth from 2 to 6 layers (ablation) | MVC text encoder · LJSpeech | pitch RMSE (Hz): 27.5 -> 23.2 | Improved |
| increase diffusion steps from 5 to 9 (ablation) | MVC · LJSpeech validation | MOS-N: 3.87 -> 3.89 (RTF 0.0177 -> 0.0221) | No clear effect |
| increase diffusion steps from 3 to 5 (ablation) | MVC · LJSpeech validation | MOS-N: 3.62 -> 3.87 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- StyleTTS2 — StyleTTS 2: Towards Human-Level Text-to-Speech through Style Diffusion and Adversarial Training with Large Speech Language Models (NeurIPS 2023) · better on 9 of 11 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- DELTA-TTS: Adapting Autoregressive Model into a Diffusion Language Model for Text-to-Speech (NeurIPS 2026 · similarity 0.89)
- Understanding Input Selectivity in Mamba: Impact on Approximation Power, Memorization, and Associative Recall Capacity (ICML 2025 · similarity 0.87)
- LongMamba: Enhancing Mamba's Long-Context Capabilities via Training-Free Receptive Field Enlargement (ICLR 2025 · similarity 0.87)
- Block-Biased Mamba for Long-Range Sequence Processing (NeurIPS 2025 · similarity 0.87)
- Can Mamba Learn How To Learn? A Comparative Study on In-Context Learning Tasks (ICML 2024 · similarity 0.86)
- CLaM-TTS: Improving Neural Codec Language Model for Zero-Shot Text-to-Speech (ICLR 2024 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
